Which AI model should I use?
Pick what you want to do. You get three models worth comparing, what each one can do, and how far a plan's credits go. Then ask all three the same question and judge the answers yourself.
Models and prices from EVA's live catalog, checked November 22, 2026.
1. What do you want to do?
Write, review and debug code.
Three models worth comparing for coding
Anthropic
about 68 credits per short message
- Thinks step by step
- Web search
- Reads images
- PDFs and files
- Creates images
OpenAI
about 34 credits per short message
- Thinks step by step
- Web search
- Reads images
- PDFs and files
- Creates images
DeepSeek
about 14 credits per short message
- Thinks step by step
- Web search
- PDFs and files
- Creates images
Tip: Give the model the whole file or function and the exact error message. For a tricky bug, compare two or three answers: they can catch different issues.
2. How far do your credits go?
| Model | Per short message | Short messages a month on Plus |
|---|---|---|
| Claude Opus 5.5 | about 68 credits | about 735 |
| GPT-6.2 Sol | about 34 credits | about 1,470 |
| DeepSeek V4 Pro 0813 | about 14 credits | about 3,571 |
| All 3 side by side (Multi Chat) | about 116 credits | about 431 comparisons |
A short message here means about 200 tokens in and 400 out (a short question and a few paragraphs back), at the model's list price. Long prompts, files, web searches, images and step-by-step thinking use more. Estimates from EVA's live catalog, not a quote. Plans and prices
By continuing, you agree to EVA's Terms of Service and acknowledge the Privacy Policy, and confirm you are 18 or older. We'll occasionally email you product tips and updates; you can unsubscribe anytime.
Models to compare, by task
The same picks as above, all on one page. Each model links to its details.
Coding
Write, review and debug code.
- Claude Opus 5.5 (Anthropic, about 68 credits per short message)
- GPT-6.2 Sol (OpenAI, about 34 credits per short message)
- DeepSeek V4 Pro 0813 (DeepSeek, about 14 credits per short message)
Give the model the whole file or function and the exact error message. For a tricky bug, compare two or three answers: they can catch different issues.
Writing and editing
Emails, posts, articles and rewrites.
- Claude Sonnet 5.5 (Anthropic, about 34 credits per short message)
- Gemini 4 Carbon (Google, about 34 credits per short message)
- Mistral Large 4 (Mistral, about 8 credits per short message)
Tone is personal, so comparing side by side is especially useful here. Paste a short sample of your own writing and ask each model to match it.
Research with live web sources
Current facts, with links you can check.
- Perplexity Pro Search (Perplexity, about 51 credits per short message)
- GPT-6.2 Sol (OpenAI, about 34 credits per short message)
- Gemini 3.8 Flash (Google, about 13 credits per short message)
Ask for sources and open them. When two models cite different sources for the same fact, that is the claim to double-check.
Long documents and PDFs
Summarize, question or compare files.
- Gemini 4 Carbon (Google, about 34 credits per short message)
- Claude Sonnet 5.5 (Anthropic, about 34 credits per short message)
- GPT-6.1 Terra (OpenAI, about 17 credits per short message)
Upload the file and ask specific questions ("What does section 4 say about refunds?") rather than "summarize this". Long files use more credits.
Creating images
Generate or edit pictures from a prompt.
- GPT-6.2 Sol (OpenAI, about 34 credits per short message)
- Gemini 4 Carbon (Google, about 34 credits per short message)
- Grok 4.7 (xAI, about 22 credits per short message)
Each model draws in its own style, so the same prompt gives different results. Check each image provider's terms before using results commercially.
Understanding images and screenshots
Read charts, photos, receipts and UI screenshots.
- Gemini 4 Neon (Google, about 51 credits per short message)
- GPT-6.2 Sol (OpenAI, about 34 credits per short message)
- Qwen3.8 Omni Flash (Qwen, about 2 credits per short message)
Crop to the part that matters and say what you need ("list every number in this table").
Video and audio
Ask questions about a clip or a recording.
- Gemini 4 Carbon (Google, about 34 credits per short message)
- Gemini 3.8 Flash (Google, about 13 credits per short message)
- Qwen3.8 Omni Flash (Qwen, about 2 credits per short message)
Short clips work best. Ask about specific moments ("what is said after the first minute?").
Hard reasoning, math and analysis
Multi-step problems where getting it right matters.
- GPT-6.2 Astra (OpenAI, about 135 credits per short message)
- Claude Fable 5.2 (Anthropic, about 135 credits per short message)
- Gemini 4 Argon (Google, about 34 credits per short message)
These models think before answering, which takes longer and uses more credits. When two of them agree, that's a good sign, not proof. Check the steps.
Everyday questions on a budget
Quick answers that use very few credits.
- GPT-6 Luna (OpenAI, about 2 credits per short message)
- Claude Haiku 5.5 (Anthropic, about 2 credits per short message)
- Gemini 3.5 Flash-Lite (Google, about 8 credits per short message)
Small models are fast and use very few credits. If an answer matters, run it past a larger model too.
Building on the API instead?
The AI cost calculator compares providers' API list prices per request and per month, for any token count.
Open the AI cost calculatorFrequently asked questions
- Which AI model is best?
- There isn't one best model for everything. Models differ in what they can do (web search, reading files, images, video), in how they write, and in price. The reliable way to choose is to send the same prompt to two or three candidates and compare the answers for your own task.
- How were these models picked?
- The EVA team keeps a short list per task. The page shows the first three from that list that are live in EVA today and support what the task needs, for example web search for research or video input for video. Prices and capabilities are read from EVA's live model catalog every hour.
- What does a credit cost estimate mean here?
- It is the cost of one short exchange (about 200 tokens in and 400 out) at the model's list price, converted to EVA credits. Longer prompts, uploaded files, web searches, images and models that think step by step use more. It is an estimate, not a quote.
- How many credits does each EVA plan include?
- Free (free) includes 5,000 credits a month; Basic ($9.90 a month or $96 a year) includes 20,000 credits a month; Plus ($19.90 a month or $192 a year) includes 50,000 credits a month; Premium ($49.90 a month or $480 a year) includes 100,000 credits a month. Every plan can use every model.
- Can I compare models side by side?
- Yes. EVA's Multi Chat sends one prompt to up to four models at once and shows the answers next to each other. A comparison uses the credits of each model in it.